Sprechen sie mit uns +43 1 505 9744 0

Machine Translation

Machine translation (abbreviated to MT) is the process of translating a text from the source language into the target language using a computer program.

Machine translation is a subfield of artificial intelligence. While human translation is a focus of applied linguistics, research into machine translation is primarily carried out in the fields of IT and computer linguistics.

Although the quality of machine translation is improving, it’s still not yet good enough to replace human translation altogether. In order to achieve the same level of quality as with human translation, the data must first be prepared (pre-editing) and after the translation, the text must be reviewed (post-editing). Today, machine translation is used primarily for real-time texts (i.e. forums, chats) where the focus is on gathering information and not on the quality of the language.

There are many varieties of machine translation. The most important are:

  • Rule-based machine translation: this is the oldest method of MT and is based on grammatical rules and linguistic terminology definitions. During the translation, the source text is analysed sentence by sentence and broken up into its grammatical components. Then, the sentence structure is transfered to the target language and, lastly, the individual words are translated and inserted into the transfered structure. Rule-based machine translation can only be applied for one particular language pair, as the grammatical rules of each language vary.
  • Statistical machine translation: this new approach does not attempt to apply the rules of a language to a source text. Instead, bilingual texts are saved and the probability of specific word combinations is calculated. Since there is no language dependency, statistical based MT can be used for different languages. However, a large amount of bilingual text is required. More information can be found here.
  • Neural machine translation: Similar to statistical MT, neural machine translation is based on the analysis of bilingual texts. In addition, an artificial neural network is established, which saves a wide variety of information for each individual word, such as its context in a sentence and various other properties. This makes neural MT more flexible than statistical MT and provides reliable results for long sentences.

Nowadays, statistical machine translation used most frequently, while the use of neural machine translation is also increasing. To obtain a good quality translation, a large amount of bilingual data must be available to “train” the engine of the MT system. The translation is then corrected by a post-editor. Most CAT tools now allow for the integration of MT plugins so that the post-editing can take place in a similar environment to that of a revision. The post-editor receives the machine translated text and must adapt it according to the content, grammar, syntax and style of the source text.

Machine Translation – yes or no?

Whether or not it is appropriate to use machine translation depends on many different factors, such as the text type, subject field, financial and time limitations, and the quality required. The positives and negatives of machine translations must be considered and compared with the necessary requirements. Therefore, the most important aspects of machine translation are as follows:

  • Cost saving: Apart from acquisition and maintenance costs, no costs are incurred from using a human translator. The cost of using a post-editor is less than that of a translator.
  • Time saving: MT allows large quantities of text to be translated in a fraction of the time it would take a human translator. However, sufficient time must be allowed to train the engine and to complete the post-editing process.
  • Availability of large bodies of text: With machine translation, large quantities of bilingual text must be used to train the engine of the MT system prior to the translation. This is necessary to achieve a quality of translation that is high enough to facilitate the post-editing process.
  • Quality: The quality of a machine translation depends on the quantity of high-quality bilingual data available and how accurately the post-editing was carried out.
  • Suitable text types: Machine translation delivers the best results for large bodies of text and standardised texts with lots of repetitions, for example with documentation, manuals or operating instructions.

To find out the extent to which machine translation actually fulfills the requirements of the professional translation market, Interlingua carried out a pilot project. Find out more here.